A benchmark of GRU and LSTM networks for short-term electric load forecasting

Abstract

Recently, electric power systems have been modernised to be integrated with distributed energy systems having intermittent characteristics. Herein, short-term electric load forecasting (STLF), which covers hour, day, or week-ahead predictions of electric loads, is a crucial piece of the modern power system puzzle whose level of complexity has become more and more sophisticated owing to incorporating microgrids and smart grids. Due to the nonlinear feature of electric loads and the uncertainties in the modern power systems, deep learning algorithms are frequently applied to STLF problem which can be described as an arduous challenge because of being affected by several impacts. In this paper, gated recurrent unit (GRU) and long short-term memory (LSTM) networks are implemented in forecasting an hour-ahead electric loads of a large hospital complex located in Adana, Turkey. Overall results belonging to the benchmark of GRU and LSTM networks for STLF revealed that employing GRU networks performed better in terms of mean absolute percentage error (MAPE) by 7.8% and computational time by 15.5% in comparison with utilising LSTM networks.